A daily crypto, finance, and AI vocabulary puzzle about collateral risk, cash-like rates, retrieval, and inference efficiency.
Onchain Credit
These terms cover overcollateralization, pricing inputs, and the point where a loan becomes unsafe.
- Collateral Ratio: The collateral ratio compares the value of pledged collateral with the value of the debt it backs.
- Overcollateralization: Overcollateralization means posting collateral worth more than the amount borrowed.
- Oracle Price: An oracle price is an off-chain market value fed into a smart contract or lending system.
- Liquidation Threshold: A liquidation threshold is the collateral level that can trigger forced closing or deleveraging.
Treasury Yields
These terms connect government debt, interest-rate sensitivity, and the spread between quoted and market prices.
Treasury BillYield CurveDurationBasis Point
- Treasury Bill: A treasury bill is a short-term U.S. government security typically sold at a discount.
- Yield Curve: A yield curve shows how interest rates vary across different maturities.
- Duration: Duration estimates how much a bond's price may change when interest rates move.
- Basis Point: A basis point is one-hundredth of a percentage point, often used to quote rate changes.
Agent Memory
These terms describe how models compress, store, and retrieve information for longer tasks.
Context WindowEmbeddingVector DatabaseRetrieval-Augmented Generation
- Context Window: A context window is the amount of input and prior output a model can consider at once.
- Embedding: An embedding turns text or other data into a numerical vector that captures meaning.
- Vector Database: A vector database stores embeddings so similar items can be retrieved efficiently.
- Retrieval-Augmented Generation: Retrieval-augmented generation combines search over external content with model generation.
Serving Throughput
These terms describe common inference techniques that trade memory, compute, and response time.
- Batching: Batching combines multiple requests so a model can process them more efficiently.
- KV Cache: A KV cache stores past attention keys and values so generation can continue faster.
- Speculative Decoding: Speculative decoding uses a smaller draft model to propose tokens before verification.
- Quantization: Quantization reduces numerical precision so models use less memory and often run faster.